Skim

A small AI tool that summarizes PDFs page-by-page based on what you tell it to look for — built for students and teachers who need to skim dense material fast.

Upload a PDF, describe what you want extracted (e.g. "extract the top points on electrolysis" or "summarize each chapter"), and Skim returns a clear, structured summary.

Live demo: https://skim.onrender.com

---

How it works

1. User uploads a PDF and types a prompt describing what they want extracted
2. Flask backend sends the PDF to Firecrawl's /parse endpoint, capped at the first 10 pages, and gets back clean markdown
3. The parsed text + user's prompt are sent to Mistral for summarization
4. The summary is returned to the frontend and displayed
5. The frontend saves the summary to localStorage so past summaries persist across sessions — no server-side storage involved

---

Stack

• Frontend: single-file HTML + Tailwind CSS (via CDN) + vanilla JavaScript — no build step, no framework
• Backend: Python + Flask
• PDF parsing: Firecrawl /parse endpoint
• Summarization: Mistral AI chat completions
• Hosting: Render (backend + static frontend served together)

The frontend deliberately skips a JS framework here — this project sits alongside other portfolio projects built with Next.js/TypeScript, so this one leans into plain HTML/CSS/JS to show range and comfort working without tooling.

---

Running locally

Backend

cd skim
python -m venv venv
source venv/bin/activate          # or venv\Scripts\activate on Windows

pip install -r requirements.txt

Create a .env file in the project root:

FIRECRAWL_API_KEY=fc-your-key-here
MISTRAL_API_KEY=your-mistral-key-here

Run the server:

python -m app.main

Visit http://localhost:5000 — Flask serves the frontend directly from static/index.html.

---

Project structure

skim/
├── app/
│   ├── main.py              # Flask app + routes
│   ├── config.py            # env vars / settings
│   └── services/
│       ├── firecrawl.py     # PDF parsing via Firecrawl
│       └── mistral.py       # summarization via Mistral
├── static/
│   └── index.html           # entire frontend — HTML, Tailwind, JS
├── requirements.txt
└── .env                     # not committed — see .gitignore

---

Known limitations

• PDFs are capped at 5MB and the first 10 pages — larger documents will be summarized partially, with a note shown in the UI
• No backend storage — the Flask API is a stateless pass-through between the frontend, Firecrawl, and Mistral
• No user accounts — summary history is per-browser via localStorage, not synced across devices
• No rate limiting — not intended for public production traffic
• Scanned/image-heavy PDFs depend on Firecrawl's OCR quality